Claude Code usage: how it works and how to check it

What counts as usage

Claude Code usage is measured against your plan in two rolling windows at once, and both have to have room for the agent to keep working.

• A session window, the short one. It opens with your first message and covers a few hours of work from that point, then resets.
• A weekly window, the long one. It is the ceiling on how much you can get through across a whole week, and it is the one that usually catches people out, because you can be well inside the session window and still be stopped.

What consumes them is tokens, and the largest consumer is usually not the message you typed. It is the context resent on every turn: a long conversation, large files read in full, and subagents doing work of their own all multiply what each reply costs. This is why two people on the same plan can hit the wall at very different points in the day.

When a window fills, Claude Code stops and tells you when it resets. Nothing is lost, and you do not need to do anything except wait or switch to a plan with more room. The exact allowances belong to your plan rather than to the CLI, so the authoritative statement of what you get is in your Anthropic account, not in any tool that reads your disk.

Work out which question you are asking

• What is this session costing me right now? Use the built-in command. It is one keystroke and it is already there.
• What have I used across every project this month? Use ccusage. It reads your local history, so it can go back as far as your session files do.
• Am I about to hit my subscription limit? That is an account question, and the account settings are the only authoritative answer.
• What is this API key spending? The Anthropic Console, not the CLI.

The built-in commands

Inside a running session, /usage reports what that session has consumed, broken down by token type. /status gives you the session's current state, including the model and the working directory, which is useful when you have several sessions open and have lost track of which is which.

These are fast and accurate for the here and now. What they will not do is aggregate: close the session and the number goes with it.

ccusage, for everything historical

Claude Code writes a JSONL transcript for every session under ~/.claude/projects, one directory per project and one file per session. Those files record the token counts. ccusage is an open-source CLI that parses them and prints reports.

Two things make it the standard answer. It needs no account, API key or network call, because the data is already on your machine. And it runs straight from npx, so there is nothing to install:

The daily view is the one most people leave running in a spare tab. The session view is the one that answers "which project is eating my quota", because it breaks the total down per session rather than per day.

Two caveats worth knowing. The costs are computed from token counts and published model pricing, so they are an accurate reconstruction rather than a bill: on a subscription plan they tell you what the same work would have cost on the API, which is a useful number but not one you are being charged. And the reports only reach as far back as the session files you still have.

Subscription limits and API spend

Plan limits are enforced by the account, not by the CLI, so no tool that only parses your local session files can authoritatively tell you how close you are; the account's own record is the one that counts. Check your Anthropic account settings for the plan you are on and what it allows. If you are running Claude Code against an API key rather than a subscription, the Anthropic Console is the place: it shows actual spend and the rate limits attached to the key, and it is the only number that matches your invoice.

If you are being cut off sooner than you expect, the usual causes are worth ruling out before blaming the limit: a very large context being resent on every turn, subagents multiplying the work, and large files read in full when a search would have done.

Where Noveriq fits, and where it does not

What it does show is the other half of the question this page opened with. Noveriq has a usage meter, on by default and switchable off in settings, that reports how much of your session window and your weekly window you have used, as a percentage with a countdown to the reset, alongside how much context each running session has left. For Claude Code that reading is not parsed from your local session files: the app calls Anthropic's own usage endpoint with your existing Claude Code login, about every five minutes, so the figure is the provider's own rather than a reconstruction. That is the same distinction this guide draws above, where a local parser can only estimate what the account knows for certain.

What it is not is a cost tracker. Noveriq is primarily a native macOS app that shows every running agent session in the MacBook notch: what each one is doing, how long it has been going, and which ones are stopped waiting for a human. You approve or deny permission requests from there, without focusing the terminal, and jump to the exact tab when you do want to be in it.

So the two sit alongside each other rather than competing: ccusage answers what it cost and what you spent over time, Noveriq answers how much window is left and whether anything needs you right now. If you are weighing them up directly, there is a page on exactly that comparison.

Questions

Does ccusage send my data anywhere?

It reads local files and prints to your terminal. Nothing about your sessions needs to leave the machine for it to work.

Why does ccusage show a cost when I am on a subscription?

It reconstructs cost from token counts and model pricing. On a subscription that figure is a comparison, not a charge.

Can I see usage for one project only?

Yes. The session view groups by session, and sessions are stored per project directory, so a single project's consumption is visible there.

Does Noveriq show token counts?

No, and no cost either. Its usage meter shows how much of the session and weekly window you have used and when each resets, read for Claude Code from Anthropic's own usage endpoint with your existing login, plus how much context each session has left. It keeps no spending history, so for tokens and dollars ccusage remains the tool.